> From: David Maxwell <david@fundy.ca>
>
> On Mon, Aug 09, 1999 at 04:56:10PM +0200, Joerg Czeranski wrote:
> > Unfortunately you're totally right.  The XL266 is ARC only.
> > So are the PC164RX boards, Samsung UX boards (I don't remember the
> > company who designed them, Samsung builds them) and API's UP2000(?)
> > NT boxes AFAIK.
> > 
> > > Any hints on how to get through the install on this
> > > particular machine?
> > 
> > You have the choice of NT or Linux.
>
> What's Linux doing that we can't do?
> (It's a trick question)

Hmm.

Perhaps what you don't know is that linux deals with ARC not by running
under it but by loading an SRM sort-of-clone into RAM and _replacing_
the ARC firmware with this ram-based layer called milo. Also, note that
linux cannot run on all arc systems, just ones for which they've written
this console replacement kludge.

> Is it just a matter of design, where ARC doesn't have some things
> that make life easy (portable?) for NetBSD?

I'm not sure I understand this question. Roughly, ARC causes the same
problems for Linux as NetBSD. On some systems. however, the old, cheap,
and very poor but free palcode that linux loads (because ARC doesn't provide
a 64-bit arch!!) is good enough for linux but not good enough for the
considerably more sophisticated NetBSD virtual memory system. However, this
is probably not the case on the XL; I imagine the palcode milo loads for
the relatively simple 21064 is good enough for us. Also, we could probably
fix palcode bugs if we cared.

Also, our VM system has been ~~ completely replaced since those test results
were seen, so it's even possible that the milo palcode would work just fine
with netbsd on all platforms these days.

> Is there more ARC information (docs) available now than when port-alpha
> started?

Yes, there is.

> Is there anything the Linux boot code couldn't show us how to handle?

No. If someone wanted to, they could make a version of milo that had our
libsa FFS code in it and deal with ARC bootblock problems. (The development
kit requires working under NT with Micro$oft compilers.) I've seen some PE
stuff in the cygnus tools, so some day it can probably done natively.

I don't want to discourage someone who wants to deal with this, I mean, it
would be nice to have, but it's a lot of work for a questionable payoff.
Of course, DEC has paid the milo authors to do most of this work for us...

If someone does want to work on this, they should contact me. There are
some pitfalls to step around...

	ross.harvey@computer.org